基于近红外的Fisher判别法鉴别废塑料

Recognition of Plastics Wastes with Fisher Discriminant Classification Based on Near Infrared Spectrum

通过对化学计量学方法的分析研究,提出了一种基于近红外光谱的丙烯腈–丁二烯–苯乙烯塑料、聚对苯二甲酸乙二酯、聚氯乙烯、聚丙烯、聚苯乙烯和聚乙烯6种塑料的一次性鉴别模型,为近红外塑料识别软件的二次开发提供有效程序。结果表明,光谱经3次多项式、13窗口S–G最小二乘拟合平滑+sym17函数、分解2层小波分析+主成分分析+特征波长选择+Fisher判别处理,可得5个判别函数式。由这些函数式建立的判别模型可以实现6种塑料的一次性识别,其校正集样本自身验证和交叉验证的识别率分别为100%和84.9%,表明该模型稳定;预测集样本进行模型外部检验的准确率为100%,表明该模型可行。

Through researching the chemometrics methods, a new method for recognition of six kinds of plastics such as ABS, PET, PVC, PP, PS and PE with Fisher discriminant classification using near infrared spectra was proposed, which provided an effective application for the secondary development of near infrared recognition software. The experimental results show that five discriminants functional expressions can be got after the spectrum processing with Savitzky-Golay smoothing in 3 polynomial and 13 windows＋the wavelet analysis in layer 2 and syml7 function＋principal component analysis＋choose characteristic wavelength ＋Fisher discriminant analysis. The recognition model which is built with these functional expressions, can be effectively used to identify the six plastics. The accuracies for its validation and cross validation are 100% and 84.9% respectively in calibration set, which shows the stability of the model. The identification accuracies for the unknown samples are 100% in test set which proves the feasibility of the model.